A multi-breakpoint methodology to detect changes in climatic time series. An application to wet season precipitation in subtropical Argentina

Abstract

Homogeneity is an important characteristic of time series that must be checked before doing any analysis. Breakpoints in meteorological time series are very commondue to climatic jumps caused by natural forcing and/or human activity but also, and mainly, produced by inhomogeneities which are erratic in nature. In this work,five breakpoint tests are analyzed to evaluate their performance in detecting breakpoints for different lengths of time series and intensity of breakpoint, among otherfeatures, through the realization of numerical experiments of sensitivity. These tests are: Student´s, Mann-Whitney, Buishand-R, Pettit and SNHT. The Student´s andMann-Whitney tests show high probability of false breakpoint detection and problems to reproduce the date when a breakpoint occurs. In addition, the Buishand-Rand Pettit are more efficient to reproduce the date of breakpoint when it occurs in the middle of a time series while the SNHT does it for breakpoints in its borders. Inthis sense, the Pettit, Buishand-R and SNHT tests show better performance than the Mann-Whitney and Student´s. Furthermore, an original methodology to detectmulti-breakpoints based on the aforementioned tests is applied to precipitation time series from sixty-two rain-gauge stations in subtropical Argentina. A breakpointaround 1976 is detected in the wet season (austral warm season), highly likely linked to the well-documented 1976/77 climate transition, by all the used tests and formost of the stations. To a lesser extent, another breakpoint occurred in the mid-1950´s. Other breakpoints are also detected in the early 1980s and the early 2000s,though in few stations and by one or two tests. For the breakpoints found in the mid-1950´s and in the early 1980´s, a strong relationship with two ENSO´s indices isfound which suggests that changes in long-term ENSO variability could be the cause of these breakpoints.